Abstract:

Genomes contain remarkable codes for controlling their
own operation but most of these codes are hidden in vast
stretches of DNA and are very hard to decipher. Fortunately,
nature's diversity allows us to explore many evolutionary
variants arising from a common ancestor, providing many versions
of similar codes to study. We are thus posed with the following
problem: given a set of evolutionary related species, each with
its own genotype (DNA) and phenotype (function, behavior), can
we explain the similarities and differences in species'
phenotypes by revealing conserved and mutated parts of the
hidden DNA regulatory codes?

In this talk, I'll describe a computational approach for the
construction of a coherent probabilistic model for genomic
regulation and its evolution. I will focus on three illustrative
examples: A) The interaction between proteins and DNA, its
surprisingly widespread quantitative and stochastic nature, and
how it affects the function and evolution of transcription
factor binding sites. B) Novel DNA codes that control the level
of DNA methylation, inferred using evolutionary analysis of
mammalian genomes. C) The organization of genes into networks of
co-regulated groups, how it slows down the evolution of their
control DNA and how evolution works around this obstacle.